[PATCH] ARM: The mandatory barrier rmb() must be a dsb() in for device accesses

[Catalin Marinas]

On Thursday, 28 April 2011, Russell King - ARM Linux
<linux at arm.linux.org.uk> wrote:
> So, what am I supposed to do with 6870/1 (rmb) and the wmb patch?
> One or other is wrong - they certainly are not both correct as rmb()
> can not be stronger than wmb().

Given Ming's reply to the wmb patch he posted, you should discard the
wmb patch. As for the rmb one (6870/1), I already explained why it is
needed.

-- 
Catalin